mirrors/icefall
1
0
Fork 0
mirror of https://github.com/k2-fsa/icefall.git synced 2025-12-11 06:55:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

Prevent left_context_chunks from being 0.

Browse Source
This commit is contained in:
Daniel Povey 2023-02-13 12:03:07 +08:00
parent dc481ca419
commit 5842de9464
1 changed files with 3 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
egs/librispeech/ASR/pruned_transducer_stateless7/zipformer.py
View File

@ -100,7 +100,7 @@ class Zipformer(EncoderInterface):
causal (bool): if True, support chunkwise causal convolution. This should
not hurt WER as no modeling power is lost, but the convolution modules will be
slightly slower and use more memory. Enables use of the chunk_size and
left_context_chunk options in forward(), which simulates streaming
left_context_chunks options in forward(), which simulates streaming
decoding.
chunk_size: (list of int): only set this to other than [-1] if causal;
the chunk size will be randomly chosen from this list. -1 means no chunking.
@ -344,6 +344,8 @@ class Zipformer(EncoderInterface):
left_context_frames = random.choice(self.left_context_frames)
# Note: in Python, -1 // n == -1 for n > 0
left_context_chunks = left_context_frames // chunk_size
if left_context_chunks == 0:
left_context_chunks = 1
return chunk_size, left_context_chunks